TXs do not re-enter mempool when a block is disconnected

[pinheadmz]

This came to my attention when a user opened bcoin-org/bcoin#915, as an amazing coincidence I was also studying the Bitcoin Core BIP68 test which shed some important light on the subject.

The issue arises when a block is disconnected (either by a reorg, or locally with rpc invalidateblock) and there are two relevant places in the code.

First: when a block is removed, if the mempool is empty, all the TXs in the block are discarded:

hsd/lib/mempool/mempool.js

Lines 333 to 339 in 27c99ad

	async _removeBlock(block, txs) {
	if (this.map.size === 0
	&& this.claims.size === 0
	&& this.airdrops.size === 0) {
	this.tip = block.prevBlock;
	return;
	}

Second, if a 'reorganize' event is caught by mempool, the mempool is "sterilized" by _handleReorg() but this function is a little brutal, it removes all CSV transactions and all coinbase-spends.

The effect of these two snippets is to simplify mempool processing after a reorg (the BIP68 rules get stupid-complicated when transactions are re-entering the mempool, especially if the parent of a CSV TX re-enters the mempool)

I don't think this is a much an issue with bcoin but it will be with Handshake, since this is the code all miners will be running, and there are more likely to be re-orgs in the early stages of the chain. The effect will be lots of transactions disappearing completely from mempools and blocks if a block is removed, the sender (or recipient) will have to re-broadcast. It is also very non-optimal for miners because they are throwing away fees.

Using the Bitcoin Core test as a guide, I'm going to try to implement deeper reorg-mempool behavior behavior in bcoin and after review and merge, port that back to hsd.